Question

我有一个scala程序，我已经实现了一个完美的二级排序。我编写该程序的方式是：

object rfmc {
  // Custom Key and partitioner

  case class RFMCKey(cId: String, R: Double, F: Double, M: Double, C: Double)
  class RFMCPartitioner(partitions: Int) extends Partitioner {
    require(partitions >= 0, "Number of partitions ($partitions) cannot be negative.")
    override def numPartitions: Int = partitions
    override def getPartition(key: Any): Int = {
      val k = key.asInstanceOf[RFMCKey]
      k.cId.hashCode() % numPartitions
    }
  }
  object RFMCKey {
    implicit def orderingBycId[A <: RFMCKey] : Ordering[A] = {
      Ordering.by(k => (k.R, k.F * -1, k.M * -1, k.C * -1))
    }
  }
  // The body of the code
  //
  //
  val x = rdd.map(RFMCKey(cust,r,f,m,c), r+","+f+","+m+","+c)
  val y = x.repartitionAndSortWithinPartitions(new RFMCPartitioner(1))
}

我想用clojure的DSL实现同样的东西，称为flambo。由于我无法使用clojure编写分区器，因此我重新使用上面的代码定义，编译它并将其用作我的Clojure代码中的依赖项。

现在我按照以下方式在clojure代码中导入分区器和键：

(ns xyz
  (:import
    [package RFMCPartitioner]
    [package RFMCKey]
    )
  )

但是当我尝试通过RFMCKey创建(RFMCKey. cust_id r f m c)时，会抛出以下错误：

java.lang.ClassCastException: org.formcept.wisdom.RFMCKey cannot be cast to java.lang.Comparable
    at org.spark-project.guava.collect.NaturalOrdering.compare(NaturalOrdering.java:28)
    at scala.math.LowPriorityOrderingImplicits$$anon$7.compare(Ordering.scala:153)
    at org.apache.spark.util.collection.ExternalSorter$$anon$8.compare(ExternalSorter.scala:170)
    at org.apache.spark.util.collection.ExternalSorter$$anon$8.compare(ExternalSorter.scala:164)
    at org.apache.spark.util.collection.TimSort.countRunAndMakeAscending(TimSort.java:252)
    at org.apache.spark.util.collection.TimSort.sort(TimSort.java:110)
    at org.apache.spark.util.collection.Sorter.sort(Sorter.scala:37)
    at org.apache.spark.util.collection.SizeTrackingPairBuffer.destructiveSortedIterator(SizeTrackingPairBuffer.scala:83)
    at org.apache.spark.util.collection.ExternalSorter.partitionedIterator(ExternalSorter.scala:687)
    at org.apache.spark.util.collection.ExternalSorter.iterator(ExternalSorter.scala:705)
    at org.apache.spark.shuffle.hash.HashShuffleReader.read(HashShuffleReader.scala:64)
    at org.apache.spark.rdd.ShuffledRDD.compute(ShuffledRDD.scala:92)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:277)
    at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:70)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:242)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:35)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:277)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:244)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:61)
    at org.apache.spark.scheduler.Task.run(Task.scala:64)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:203)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
    at java.lang.Thread.run(Thread.java:745)

我的猜测是它无法找到我在分区后定义的顺序。但如果它在Scala中有效，为什么它在Clojure中不起作用？

Answer 1

所以我终于把它弄清楚了。我必须基本上将自定义排序函数编写为单独的scala项目，然后在clojure中调用它。

我以这种方式编写了我的scala文件：

import org.apache.spark.Partitioner
import org.apache.spark.rdd.RDD

case class RFMCKey(cId: String, R: Double, F: Long, M: Double, C: Double)
class RFMCPartitioner(partitions: Int) extends Partitioner {
  require(partitions >= 0, "Number of partitions ($partitions) cannot be negative.")
  override def numPartitions: Int = partitions
  override def getPartition(key: Any): Int = {
    val k = key.asInstanceOf[RFMCKey]
    k.cId.hashCode() % numPartitions
  }
}
object RFMCKey {
  implicit def orderingBycId[A <: RFMCKey] : Ordering[A] = {
    Ordering.by(k => (k.R, k.F * -1, k.M * -1, k.C * -1))
  }
}

class rfmcSort {
  def sortWithRFMC(a: RDD[(String, (((Double, Long), Double), Double))], parts: Int): RDD[(RFMCKey, String)] = {
    val x = a.map(v => v match {
                case (custId, (((rVal, fVal), mVal),cVal)) => (RFMCKey(custId, rVal, fVal, mVal, cVal), rVal+","+fVal+","+mVal+","+cVal)
            }).repartitionAndSortWithinPartitions(new RFMCPartitioner(parts))
    x
  }
}

我将它编译为ascala项目并以这种方式在我的clojure代码中使用它：

(:import [org.formcept.wisdom rfmcSort]
         [org.apache.spark.rdd.RDD])

sorted-rfmc-records (.toJavaRDD (.sortWithRFMC (rfmcSort.) (.rdd rfmc-records) num_partitions))

请注意我从我创建的sortWithRFMC对象调用rfmcSort函数的方式。此处需要注意的一件非常重要的事情是，当您将JavaPairRDD传递给scala函数时，必须首先通过调用spark RDD方法将其转换为普通.rdd。然后你必须将spark RDD转换回JavaPairRDD以便在clojure中使用它。

Secondar使用clojure / flambo在spark中排序

1 个答案: